If so, the most important problem is to find more winners. Minority of projects are very successful, recouping costs for a larger number ofįailures. We also expect Prodigy to reduce costsįor larger projects, but it’s the increased agility we’re most excited about.ĭata science projects are said to have uneven returns, like start-ups: a Planning meeting could even be scheduled. Ideas can be tested faster than the first Prodigy solves this problem by letting data scientists conduct their ownĪnnotations, for rapid prototyping. Hosted component is adding value, instead of removing it. Solutions to these problems could surely be developed –īut… why? As attractive as SaaS is to investors, it only makes sense if the Model must be updated during the annotation session, and the updates must be The model, so that more informative examples can be chosen for annotation. The net result is a lot of wastedĪctive learning works best when you have a lot of raw input to stream through Process relies on accurate upfront planning. Learning is an inherently uncertain technology, but the waterfall annotation Models will be required for the features you’re trying to build. To produce the annotation manuals, you need to know what statistical The experiments can’t begin until the first batch of annotations areĬomplete, but the annotation team can’t start until they receive the annotation Typical approach to annotation forces projects into an uncomfortable waterfall Prodigy addresses the big remaining problem: annotation and training. New models are less than 1/10th the size. Made sure models could be installed via pip. Of those insights have been used to make spaCy better: AI DevOps was hard, so we
What’s driving success and failure for language understanding technologies. Into the most popular library of its type, giving us a lot of insight into We’ve been working on Prodigy since we first launched Explosion last year,Īlongside our open-source NLP library spaCy and ourĬonsulting projects (it’s been a busy year!). Prodigy, a downloadable tool for radically efficient machine – the learner itself is really just a compiler. This is good, because the examples are how you program the behaviour
#Prodigy app teacher data code
The code but hard to reuse the data, so building AI mostly means doingĪnnotation. Machine learning systems are built from both code and data.